Study on Mixed-Effects Models for Biomass Estimation of Cunninghamia lanceolata Plantations Based on ULS Data

Based on field-measured biomass data from 105 sample plots across six counties/county-level cities in Guangdong Province and point-cloud metrics derived from unmanned aerial vehicle laser scanning (ULS), principal component analysis (PCA) was first performed on 46 height-related variables, and the first principal component (PC1) was used to construct a stand-height principal component (H_PCA). Elastic Net was then applied to select key predictors from candidate point-cloud metrics. Four models, including a generalized linear model (GLM), generalized linear mixed-effects model (GLMM), generalized additive model (GAM), and generalized additive mixed model (GAMM), were established, with county-level random effects incorporated into the GLMM and GAMM. Model performance was evaluated using full-data fitting and five-fold cross-validation, and the intra-class correlation coefficient (ICC) was used to quantify the contribution of county-level variation. The results showed that the combination of PCA and Elastic Net effectively selected key predictors, including the stand-height principal component (H_PCA), leaf area index (LAI), and density metric D₁, thereby improving model stability and interpretability. Among all models, the GAMM performed best, with full-data fitting accuracy of R² = 0.889, RMSE = 21.362 t/hm², and MAE = 16.068 t/hm², and five-fold cross-validation accuracy of R² = 0.831, RMSE = 23.818 t/hm², and MAE = 17.683 t/hm², outperforming the other models. The ICC of the GLMM was 0.243, indicating that approximately 24.3% of the variation in biomass could be attributed to inter-county differences, and that introducing county-level stratification could substantially improve the explanatory power and predictive accuracy of the model. Overall, the integration of PCA and Elastic Net provides an effective strategy for simplifying the point-cloud feature system. Moreover, incorporating county as a random effect in mixed-effects models helps characterize county-level composite differences and improves the accuracy and robustness of regional-scale biomass estimation for Chinese fir plantations.